Swenson can explain better than me, but all USB and ethernet data is transported as data packets, not continuous streams (nature of the beasts). The "pulse" of these data packets creates what is called packet noise, which messes up the PHY and PDN of the next leg in the relay (USB receiver interface), inducing them to create their own packet noise and so the relay baton is passed, even as the old noise dies at each changeover leg.
The trick to stop the baton passing is to design a proper interface that does not create new packet noise and for this you need a few things that lead to high signal integrity (SI): low jitter reclocking, clean power distribution network (PDN), clean power/galvanic isolation, impedance matching via the cable low overall impedance of the Dac. That is what is known so far from experimentation.
There may be other ways to get to high SI and I suspect Lukasz did the same with the digi-Lampizator circuit in his SBox based Transport, where the Spdif signal was run thru a digi-Lampi tube buffer and he showed how the irregular digital incoming signal was "squared" up on an oscilloscope. He has no scientific explanation, just speculation, but the consensus was clear that SQ was significantly improved, as you would expect from perfectly square digital signals. I asked Swenson about this, but no answer. LoL